Cover

The design and development of the easyBIM package

R for HTA Workshop

28th June, 2024

Agenda

What are BIMs?

  • Budget impact analysis estimates the financial consequences of adopting new interventions within a specific healthcare setting.
  • Purpose of budget impact analysis:
    • Necessary for reimbursement
    • Communicating value to stakeholders

How are BIMs usually built?

  • BIMs are typically built in Excel. Either as a standalone model, or in the same workbook as the CEM

  • In some cases, such as when the BIM is intended to be used as a field tool to communicate value to clinicians, an engaging and user-friendly interface can be important.

Example budget impact model structure:

Section 1: Introduction to easyBIM

Why we built easyBIM: expected use cases

easyBIM is intended to be an easy-to-use R package for health economic budget impact analysis.

Section 1: Introduction to easyBIM

BIMs tend to be relatively simple models, making them an ideal R learning opportunity.

BUILDING A PACKAGE

  • Building packages was a hot topic at last year’s R for HTA workshop
  • One takeaway we had from last year was that if you have a well-documented, and well-structured R project, then you’re already not far off from a package.
  • Some of the benefits of building a package include:
    • Code Organization and Reusability
    • Collaboration and Sharing
    • Documentation

PERSONAL MOTIVATIONS

EXPLORING DIFFERENT PACKAGES

  • We hoped to use this project as a way of exploring packages and systems, and to learn best-practices for developing R packages.

  • Packages we experimented with included:

    • S7
    • cli (for improving the print() output for the classes in our package)

Section 1: Introduction to easyBIM

Agenda

easyBIM capabilities


Elements Features identified in reviewed BIMs easyBIM
Epidemiology Patient funnel ✔️
Distinction between incident and prevalent populations.
E.g., separate dosage schedule for prevalent populations vs incident populations.		 ✔️
Subgroups which determine dosage schedules and treatment eligibility. ✔️
Acquisition costs Titration ❌
RDI/Compliance ✔️
MoM used to estimate distribution across different dose sizes ✔️
Estimate optimal (cost-minimizing) combination of vials ✔️
Dose bands and vial combination tables ✔️
Include both monotherapies and joint-therapies ✔️
Administration costs Variable costs over time horizon ❌
Costs applied either per dose or per dispense ✔️
Other costs Monitoring, health state costs; AEs; premedication; subsequent treatment ✔️
Costs where the frequency changes across years ❌
Sensitivity analysis OWSA; scenario analysis ❌

Section 2: Methods

Building a basic BIM

  • In the following section, we’ll walkthrough each of the steps required to build a simple BIM with the following features:
    • Includes acquisition costs, administration costs, monitoring costs and treatment-related-adverse-events (TRAEs)
    • Multiple formulations for each treatment
    • Treatment specific time-on-treatment (ToT)

Necessary arguments in run_model()

# Run model
base_case <- easyBIM::run_model(
  currency = "$",
  time_horizon = 5,
  default_time_interval = "monthly",
  patient_weight = list(mean = 80),
  patient_funnel = patient_funnel_instance,
  market_shares = market_shares,
  bim_treatment_list = bim_treatment_list,
  list_admin_cost = list_admin_cost
)

Note

  • In this example we won’t go through all features of easyBIM. For example, we won’t cover:
    • Vial combination optimisation
    • Methods of moment distribution of dose size dependent on patient characteristics such as patient weight and BSA
    • Subgroup analysis

Section 2: Methods

Example BIM: Creating a simple patient funnel

# Create a vector 
funnel_proportions <- c("Overall Prevalence" = 0.005,
                         "Proportion diagnosed" = 0.75,
                         "Proportion adults" = 0.2143,
                         "Proportion treated" = 0.7)

# Create an instance of the patient_funnel class
patient_funnel_instance <- easyBIM::patient_funnel(
  total_population = 67000000,
  funnel_proportions = funnel_proportions,
  annual_population_growth = 0.004
)

patient_funnel_instance

Console output

── Patient funnel ────────────────────────────────────────────────────────────────────────────────────────────
ℹ If you wish to use your own patient funnel, you can directly input the number of patients treated by defining the optional argument `number_patients_treated` in `easyBIM::run_model()` .
── Patient funnel ──
→ Total population size: 67,000,000
                     Proportions     Count
Overall Prevalence        0.500% 335000.00
Proportion diagnosed     75.000% 251250.00
Proportion adults        21.430%  53842.88
Proportion treated       70.000%  37690.01
── Number of patients treated in each year of time horizon: ──
  year_1   year_2   year_3   year_4   year_5 
37690.01 37992.14 38144.10 38296.68 38449.87
  • easyBIM includes functionality to build a basic patient funnel.
  • In some cases, patient funnels can be considerably more complex, especially when considering subgroups and treatment lines.
  • For this reason, number_patients_treated has been included as an optional argument in run_model().
    • This enables the user to use their own patient funnel and overwrite the number of patients treated in each of year of the time horizon with their own values.

Add to run_model():

# Run model
base_case <- easyBIM::run_model(
  currency = "$",
  time_horizon = 5,
  default_time_interval = "monthly",
  patient_weight = list(mean = 80),
  patient_funnel = patient_funnel_instance,
  market_shares = market_shares,
  bim_treatment_list = bim_treatment_list,
  list_admin_cost = list_admin_cost
)

Section 2: Methods

Example BIM: Market share inputs

# Create a list for both scenarios
market_share_new <- list(
  year_1 = c(0.2, 0.3, 0.5), 
  year_2 = c(0.25, 0.3, 0.45),
  year_3 = c(0.3, 0.3, 0.4),
  year_4 = c(0.4, 0.3, 0.4),
  year_5 = c(0.5, 0.3, 0.4)
)

market_share_ref <- list(
  year_1 = c(0, 0.35, 0.65),
  year_2 = c(0, 0.35, 0.65),
  year_3 = c(0, 0.35, 0.65),
  year_4 = c(0, 0.31, 0.65),
  year_5 = c(0, 0.31, 0.65)
)

market_shares <- list(market_share_new = market_share_new,
                      market_share_ref = market_share_ref)

Warning

! Market shares must sum to 1 for each year.
ℹ Market shares in list_market_share_new for years, "4, 5", were normalized to sum to 1.
ℹ Market shares in list_market_share_ref for years, "4, 5", were normalized to sum to 1.
→ To see the updated market shares, see market_shares (an output of `easyBIM::run_model()`)
  • There are 3 treatments included in this example model.
  • Market shares are inputted as lists: a list for the scenario with the new intervention, and a list for the scenario without the new intervention

Add to run_model():

# Run model
base_case <- easyBIM::run_model(
  currency = "$",
  time_horizon = 5,
  default_time_interval = "monthly",
  patient_weight = list(mean = 80),
  patient_funnel = patient_funnel_instance,
  market_shares = market_shares,
  bim_treatment_list = bim_treatment_list,
  list_admin_cost = list_admin_cost
)

Section 2: Methods

Example BIM: Creating bim_treatment objects

  • easyBIM::bim_treatment is an S7 class. It essentially performs like a list with some differences:
    • Custom print() method
    • Property classes are automatically validated
      • E.g., bim_treatment@basis is set to class_character. If the user makes the input a list, it will return an error.


bim_treatment inputs for the comparator cmp_1 + cmp_2

tx_1 <- bim_treatment(name = "cmp_1 + cmp_2",
                      components = c("cmp_1", "cmp_2"),
                      list_price =  list("cmp_1" = c(100, 150, 200),
                                         "cmp_2" = c(15, 20, 25)),
                      units_per_pack = list("cmp_1" = c(1, 1, 1),
                                            "cmp_2" = c(1, 1, 1)),
                      mg_per_unit = list("cmp_1" = c(40, 60, 70),
                                         "cmp_2" = c(50, 60, 70)),
                      ToT = 2.41,
                      dose_size = c(1, 3),
                      basis = c("mg/kg", "mg"),
                      admin_route = c("immuno_admin", "immuno_admin"),
                      frequency = c("Q3W", "Q2W")
)

Console output

── bim_treatment object: cmp_1 + cmp_2 ───────────────────────────────────────────────────────────────────────
── Acquisition unit costs ──
ℹ Each element of the vectors below corresponds to a different formulation.
               cmp_1       cmp_2   
list_price     100,150,200 15,20,25
units_per_pack 1,1,1       1,1,1   
mg_per_unit    40,60,70    50,60,70
── Acquisition and administration costs summary ──
                       cmp_1               cmp_2              
dose_size              1                   3                  
basis                  mg/kg               mg                 
frequency              Q3W                 Q2W                
admin_route            immuno_admin        immuno_admin       
time_on_treatment      2.41                2.41               
cost_per_dose          Not calculated yet. Not calculated yet.
num_doses_per_year     Not calculated yet. Not calculated yet.
acq_cost_per_patient   Not calculated yet. Not calculated yet.
admin_cost_per_patient Not calculated yet. Not calculated yet.
── Vial combinations ──
Not calculated yet.

Section 2: Methods

Example BIM: Creating bim_treatment objects

  • easyBIM::bim_treatment is an S7 class. It essentially performs like a list with some differences:
    • Custom print() method
    • Property classes are automatically validated
      • E.g., bim_treatment@basis is set to class_character. If the user makes the input a list, it will return an error.


tx_2 <- bim_treatment(name = regimen_names[2],
                      components = c("cmp_1", "cmp_2"),
                      list_price =  list("cmp_1" = c(100, 150, 200),
                                         "cmp_2" = c(15, 20, 25)),
                      units_per_pack = list("cmp_1" = c(1, 1, 1),
                                            "cmp_2" = c(1, 1, 1)),
                      mg_per_unit = list("cmp_1" = c(40, 60, 70),
                                         "cmp_2" = c(50, 60, 70)),
                      ToT = 2.41,
                      dose_size = c(1, 3),
                      admin_route = c("immuno_admin", "immuno_admin"),
                      basis = c("mg/kg", "mg"),
                      frequency = c("Q3W", "Q2W")
)
  • Warnings and error catches like the below help with transparency and show the user additional functionality to consider using.

Warning

! The same @ToT (2.41) is assumed to apply to all treatments in regimen cmp_1 + cmp_2.
→ For treatment specific inputs, include a @ToT input for each treatment in the regimen.

Add to run_model():

# Run model
base_case <- easyBIM::run_model(
  currency = "$",
  time_horizon = 5,
  default_time_interval = "monthly",
  patient_weight = list(mean = 80),
  patient_funnel = patient_funnel_instance,
  market_shares = market_shares,
  bim_treatment_list = bim_treatment_list,
  list_admin_cost = list_admin_cost
)

Section 2: Methods

Example BIM: Administration costs

Administration cost inputs

list_admin_cost <- list(
  immuno_admin = c(306.90, "per_dose"),
  oral = c(24, "per_dispense")
)
  • Administration costs
    • Administration costs can be applied each dose (per_dose), or per dispense (per_dispense), which is set at once per month.
    • Next steps: allow for a user-defined frequency.

Add to run_model():

# Run model
base_case <- easyBIM::run_model(
  currency = "$",
  time_horizon = 5,
  default_time_interval = "monthly",
  patient_weight = list(mean = 80),
  patient_funnel = patient_funnel_instance,
  market_shares = market_shares,
  bim_treatment_list = bim_treatment_list,
  list_admin_cost = list_admin_cost
)

Section 2: Methods

Example BIM: Results

print(base_case)

Summary of inputs

  • Acquisition cost inputs
  • Dosage schedule inputs
  • Intermedirary outputs (e.g. cost per dose)

Results

Plots

Console output

! The same @ToT (2.41) is assumed to apply to all treatments in regimen cmp_1 + cmp_2.
→ For treatment specific inputs, include a @ToT input for each treatment in the regimen.
! Market shares must sum to 1 for each year.
ℹ Market shares in list_market_share_new for years, "4, 5", were normalized to sum to 1.
ℹ Market shares in list_market_share_ref for years, "4, 5", were normalized to sum to 1.
→ To see the updated market shares, see market_shares (an output of `easyBIM::run_model()`)
── Model Inputs ──────────────────────────────────────────────────────────────────────────────────────────────
── Acquisition Cost Inputs: Pack Costs test ──
   regimen_name treatments  list_price units_per_pack mg_per_unit
1     Product Y  Product Y           5              1        2000
2 cmp_1 + cmp_2      cmp_1 100,150,200          1,1,1    40,60,70
3 cmp_1 + cmp_2      cmp_2    15,20,25          1,1,1    50,60,70
4         cmp_3      cmp_3        2633              1         240
── Dosage schedule inputs: ──
   regimen_name components dose_size basis frequency  ToT  admin_route vial_sharing
1     Product Y  Product Y        50 mg/kg       Q1W 4.50 immuno_admin         TRUE
2 cmp_1 + cmp_2      cmp_1         1 mg/kg       Q3W 2.41 immuno_admin         TRUE
3 cmp_1 + cmp_2      cmp_2         3    mg       Q2W 2.41 immuno_admin         TRUE
4         cmp_3      cmp_3       240    mg       BID 7.77         oral         TRUE
── Intermediary Outputs: Drug Costs ──
   regimen_name components cost_per_dose doses_per_ToT acq_cost_per_patient admin_cost_per_patient
1     Product Y  Product Y        $10.00     19.566964                 $196              $6,005.10
2 cmp_1 + cmp_2      cmp_1       $200.00      3.493065                 $699              $1,072.02
3 cmp_1 + cmp_2      cmp_2         $0.90      5.239598                   $5              $1,608.03
4         cmp_3      cmp_3     $2,633.00    472.998750           $1,245,406                $186.48

Section 2: Methods

Example BIM: Results

print(base_case)

Summary of inputs

Results

  • Cost per patient, for each cost category
  • Number of patients treated
  • Total costs and budget impact
  • Cost breakdown by category
  • Cost breakdown by sub-treatment and category

Plots

Console output

! The same @ToT (2.41) is assumed to apply to all treatments in regimen cmp_1 + cmp_2.
→ For treatment specific inputs, include a @ToT input for each treatment in the regimen.
── Model Results ─────────────────────────────────────────────────────────────────────────────────────────────
── Budget Impact Summary Results ──
                                                          Year 1          Year 2           Year 3
Eligible population for treatment with Product Y        37690.01        37992.14          38144.1
Population expected to receive Product Y                    7538         9498.03         11443.23
Total cost: without Product Y                    $30,894,495,965 $30,886,607,341  $31,005,095,862
Total cost: with Product Y                       $23,892,903,630 $21,474,533,899  $19,191,600,358
Net budget impact                                -$7,001,592,335 -$9,412,073,442 -$11,813,495,504
── Costs per patient ──
        Regimen acquisition_cost Admin_Cost monitoring_costs_PFS   TRAEs      total
1 cmp_1 + cmp_2             $703  $2,680.05               $2,100  $54.60     $5,538
2         cmp_3       $1,245,406    $186.48               $1,900 $122.00 $1,247,614
3     Product Y             $196  $6,005.10               $8,850  $30.70    $15,081
── Annual cost breakdown ──
── Scenario with new intervention 
   Year acquisition_costs  admin_costs new.monitoring_costs_PFS  new.TRAEs           total
 Year 1   $23,479,105,864  $79,084,072             $333,556,611 $1,157,083 $23,892,903,630
 Year 2   $21,301,904,941  $90,771,102              $79,783,485 $2,074,371 $21,474,533,899
 Year 3   $19,012,241,488 $102,231,492              $72,473,798 $4,653,581 $19,191,600,358
── Scenario without new intervention 
   Year acquisition_costs admin_costs ref.monitoring_costs_PFS  ref.TRAEs           total
 Year 1   $30,519,859,839 $39,922,432             $333,556,611 $1,157,083 $30,894,495,965
 Year 2   $30,764,507,035 $40,242,450              $79,783,485 $2,074,371 $30,886,607,341
 Year 3   $30,887,565,063 $40,403,420              $72,473,798 $4,653,581 $31,005,095,862
── Additional outputs ────────────────────────────────────────────────────────────────────────────────────────
[1] "See 'yearly_cost_breakdown' in the object returned by 'run_model' for a breakdown of each cost-category per regimen"

Section 2: Methods

Example BIM: Results

print(base_case)

Summary of inputs

Results

Plots

  • Market shares
  • Budget impact over time

Section 2: Methods

Agenda

Learnings and reflections

S7

easyBIM doesn’t yet make full use of the potential benefits of S7.

S7 from a developer perspective:

  • ➕ Intuitive Syntax: Simplifies the process of defining classes and methods.
  • ➖ S7 is still experimental and is not yet part of base R, so substantial future changes can occur.

S7 from a user perspective:

  • ➕ Print methods (although this is a benefit of classes in general, not specific to S7).
  • ➕ Error Prevention: S7 enforces type checks, ensuring that properties are set correctly.
  • ➕ Documentation: S7 classes inherently provide information about their structure and properties.

Functionality of easyBIM

Ease of use:

  • ➕ Predefined functions simplify user interaction, reduce errors, and provide quick results for common scenarios.

  • ➖ Dependency on Assumptions: Predefined functions rely on assumptions built into the package.




Flexibility:

  • Customization options allow adaptation to unique user needs.
  • ➖ Complexity: Increased flexibility often comes with a steeper learning curve.
  • ➖ Requires comprehensive documentation to guide users in using flexible features effectively.

Section 3: Reflections and next steps

Next steps

Section 3: Reflections and next steps

Appendix: Additional cost categories

Additional cost category inputs

# In this example, there are costs which should only apply 
# while the patients are in the progression-free-survival state.
median_PFS <- list("PRODUCT Y" = 3,
                   "cmp_1 + cmp_2" = 2,
                   "cmp_3" = 1)

# Repeated costs (e.g. monitoring costs)
inputs_monitoring_costs_PFS <- repeated_cost(
  element_names = c("Inpatient visit",
                    "Hospitalization",
                    "Outpatient visit"),
  # It is not necessary to name each element of the vectors below:
  # Names have been added for demonstration purposes.
  frequency = list("PRODUCT Y" = c("Inpatient visit" = 0, 
                                   "Hospitalization" = 1, 
                                   "Outpatient visit" = 2), 
                   "cmp_1 + cmp_2"  = c("Inpatient visit" = 0, 
                                        "Hospitalization" = 0, 
                                        "Outpatient visit" =3),
                   "cmp_3"= c("Inpatient visit" = 0, 
                              "Hospitalization" = 2, 
                              "Outpatient visit" = 0)),
  time_interval = "monthly",
  num_intervals = median_PFS, # Number of time intervals
  unit_cost = c(150, 190, 70) # Per event
)
  • Additional cost categories
    • easyBIM includes a class ‘repeated_cost’ which can be used to include costs such as monitoring costs and TRAEs. Combining costs with frequency over a specified time period.

Add to run_model():

# Run model
base_case <- easyBIM::run_model(
  currency = "$",
  time_horizon = 5,
  default_time_interval = "monthly",
  patient_weight = list(mean = 80),
  patient_funnel = patient_funnel_instance,
  market_shares = market_shares,
  bim_treatment_list = bim_treatment_list,
  list_admin_cost = list_admin_cost,
  additional_cost_categories = additional_cost_categories
)

Section 2: Methods